Skip to content

Customer Inquiry → Next Action

Can AI interpret an incomplete customer inquiry, identify what is missing, and prepare a sensible next step?

Published
15 Aug 2026
Type
Exploratory workflow comparison
Sample
10 cases
Author
FIELD

Summary

We ran 10 synthetic customer inquiries through a qualification workflow — interpret the message, identify missing information, decide the next action — and compared how a human and an AI model handled them. The inquiries were incomplete more often than not, and the next action was rarely obvious.

Research question

When a customer inquiry arrives incomplete, can AI interpret it, separate what is known from what is unknown, and prepare an appropriate next action?

Method

Ten synthetic customer inquiries were written to resemble the kind of messages a small business receives. Each inquiry was read and processed twice: once by a human working intuitively, and once by an AI model instructed to interpret the message, list what was known and unknown, and propose a next action. Outputs were compared qualitatively. No timing or scoring instrument was used.

Workflow

  1. 01Customer inquiry arrives
  2. 02Interpret intent and urgency
  3. 03Identify missing information
  4. 04Decide the next action

Dataset

Ten synthetic customer inquiries. Synthetic means the messages were written for the experiment, not collected from real customers. They were designed to vary in completeness, intent and urgency.

Baselines

Human baseline

A human read each inquiry and decided the next action the way an operator normally would: quickly and largely intuitively. The human reasoning was not written down in a standard format, which itself became an observation.

AI baseline

The AI model was given the same inquiries and asked to interpret the request, state what information was present, state what was missing, and propose a next action. The output was reviewed by a human.

Evaluation criteria

Interpretation
Did the output correctly understand what the customer was asking for?
Missing information
Were the gaps in the inquiry identified?
Next action
Was the proposed next step reasonable given what was and was not known?

Results

Results are qualitative. Most inquiries were missing information needed to act. Intent and urgency varied and were not always visible from the wording. Where the AI separated known from unknown before proposing an action, the proposed action tended to be more defensible. Human handling was faster to judge but less consistent in how the judgment was expressed.

Key findings

  • Inquiries are often incomplete; missing information is a normal condition, not an exception.
  • Intent and urgency can differ — a message can be urgent without high intent, or high intent without urgency.
  • The next action is often not obvious from the inquiry text alone.
  • Context changes meaning: the same sentence can call for different actions in different situations.
  • A price question does not necessarily mean high buying intent.
  • AI output was more useful when it explicitly distinguished known from unknown before acting.
  • Qualification can be structured into repeatable fields (request, key details, missing information, action).
  • Human judgment was often intuitive rather than standardized.

Show the failure

Failure modes

  • F-ASSUMEAssumed intent

    Treating a price question as a purchase decision.

  • F-GAPUnflagged gaps

    Proposing an action without stating what was still unknown.

  • F-CONTEXTContext loss

    Reading the message literally when context suggested a different meaning.

Observed → Inferred → Unknown

Observed

  • Most of the ten inquiries lacked at least one detail needed to quote or act.
  • Some inquiries expressed urgency without stating what was actually required.
  • The human decided quickly, but the reasoning behind each decision was not consistent from case to case.
  • When asked to separate known from unknown, the AI produced a repeatable field structure across all ten cases.

Inferred

Interpretations — our reading of the evidence, labelled as such.

  • AI's useful business role here may be standardization rather than automation: making the qualification step consistent, not removing the human from it.
  • Missing information appears to be as important as stated information when deciding the next action.
  • Human judgment and AI may differ in how they interpret the same inquiry, and neither reading was always the better one.

Unknown

What this experiment does not establish.

  • Whether the same pattern holds with real (non-synthetic) inquiries.
  • Whether the AI's next actions would actually produce better customer outcomes.
  • How much time, if any, the AI step saves in practice — time was not measured.
  • Which AI configuration or model matters; this was not a tool comparison.

What we learned

This experiment produced the first version of the repeatable qualification structure (request, key details, missing information, action) that Experiment #002 later used as its output format.

Limitations

  • Ten synthetic cases written by the researcher — not real customer messages.
  • Single human baseline; one person's judgment.
  • No timing, scoring or blind evaluation.
  • Qualitative comparison only. Nothing here establishes a controlled result.

Practical implications

What operators might reasonably take from this evidence.

  • Treat missing information as a first-class part of the inquiry, not an annoyance.
  • If using AI in inquiry handling, ask it to state known and unknown before proposing an action.
  • Do not read a price question as a buying signal by default.
  • Keep a human in the loop for the action itself; use AI for the preparation.

Conclusion

In this experiment, incomplete customer inquiries were the norm, and the most useful thing the AI did was make the missing information visible before proposing an action. That is a modest role — structured preparation, not autonomous handling — but it is a real one. Whether it generalizes beyond ten synthetic cases is an open question.

Findings from this experiment

Related experiments

Research updates

Follow the research

Get new FIELD experiments and findings when we publish them.

Email only. No lead magnet. Unsubscribe any time.